UK's 'Wait-and-See' AI Copyright Stance Risks Innovation Lag

On 18 March 2026, the UK government formally announced it would shelve planned reforms to copyright exceptions for AI training, instead prioritising a voluntary licensing and transparency framework. The decision—outlined in a Department for Science, Innovation and Technology (DSIT) position paper—marks a significant divergence from the regulatory certainty being established across the Atlantic and the EU, where copyright frameworks for generative AI are already taking shape.

For Chief AI Officers and enterprise technology leaders, the implications are substantial. While the government frames its approach as pragmatic and innovation-friendly, analysis from legal scholars, industry bodies, and competitive intelligence suggests the UK risks ceding strategic advantage to firms operating under clearer regulatory regimes.

This article examines the government's reasoning, compares the UK approach to international precedents, and assesses the real-world impact on UK AI competitiveness.

The March 2026 Decision: What Changed?

The UK government's intellectual property policy team had spent 18 months consulting on potential copyright exceptions for AI training—a common practice in jurisdictions facilitating machine learning on copyrighted materials. The approach was initially aligned with emerging global thinking: enable AI developers to train on copyrighted text, images, and video without prior licensing, subject to transparency and opt-out mechanisms.

On 18 March, DSIT published updated guidance on AI and copyright, confirming the government would not pursue statutory exceptions. Instead, it would:

• Monitor EU and US regulatory developments for 12-18 months
• Encourage voluntary industry licensing agreements
• Strengthen transparency requirements for AI model training data
• Task the Alan Turing Institute with evidence-gathering on licensing effectiveness
• Review the decision in Q1 2027

The decision surprised many in the AI and creative industries, who had expected clearer statutory direction. The UK AI regulation approach has historically favoured principles-based, light-touch governance. This copyright decision extends that philosophy but risks leaving a critical gap at a moment when competitors are establishing binding rules.

Why the Government Chose Delay

Three strategic rationales emerge from DSIT's position paper and subsequent policy announcements:

Licensing as a Market Solution

DSIT argues that voluntary licensing—already emerging through partnerships between AI labs and publishers—can deliver equivalent outcomes to statutory exceptions without the regulatory burden. Industry bodies including the Publishers Association and the Copyright Licensing Agency (CLA) have signalled support for licensing frameworks, viewing them as revenue-generating alternatives to statutory carve-outs.

This reflects broader UK AI policy philosophy: enable market mechanisms before imposing rules. The logic is sound in principle—licensing aligns incentives and allows creators to be compensated—but depends on genuine competitive pressure to licence fairly. Early evidence suggests the market is not yet delivering that pressure uniformly.

International Regulatory Divergence

The EU AI Act, implemented in phases from 2024 onwards, established a risk-tiered approach but deferred detailed copyright guidance to 2025-2026 negotiations. The US, through Congressional pressure and USPTO guidance (2024-2025), has signalled scepticism toward AI-specific copyright exceptions, instead emphasising fair use doctrine as sufficient for training activities.

By pausing, the UK government signals it will align with whichever international standard emerges as dominant. This hedging strategy avoids being the first mover if the EU retreats from exceptions, but also risks being last if exceptions become a competitive advantage.

Creative Industry Lobbying

The music, publishing, and visual arts sectors have been vocal opponents of AI copyright exceptions. The UK music industry, through the BPI and PPL, has lobbied intensively against exceptions that would allow AI labs to train on copyrighted recordings without licensing. Authors' groups including the Society of Authors have similarly opposed text-training exceptions.

The government's decision reflects this pressure. However, it also reveals a structural tension: creative industries have genuine revenue concerns, but opt-out-based exceptions (already adopted in some EU member states) would address those concerns while enabling training innovation.

International Comparison: Where Others Stand

European Union

The EU AI Act establishes a phased approach, with copyright guidance emerging across 2025-2026. Key points:

• Text and data mining (TDM) exceptions: The 2019 Directive on copyright in the Digital Single Market permits TDM for research organisations without needing to seek explicit permission, though commercial use requires licensing negotiations with rights holders.
• AI-specific review: The European Commission's AI Office is conducting a three-year evidence review on copyright exceptions for high-risk AI systems. Final guidance is expected Q4 2026.
• Member state divergence: France and Germany are pursuing opt-out frameworks where AI developers can train on copyrighted content unless creators explicitly object. This approach is gaining traction as a compromise between exception and licensing.

The EU's approach is slower than some expected, but it provides legal clarity through multiple pathways. UK firms operating across the EU now face regulatory fragmentation—something the UK hoped to avoid post-Brexit.

United States

The US regulatory stance is more permissive than the UK is currently signalling:

• Fair use doctrine: US courts have consistently upheld AI training as fair use under copyright law, particularly where training is transformative and non-commercial or indirect commercial use. The 2024 Thaler v. USPTO ruling reinforced broad interpretation of fair use.
• Congressional stance: Proposed AI regulation (e.g., the AI Regulatory Authority Act of 2025) does not include copyright exceptions but implicitly accepts fair use as sufficient. This gives US labs substantial freedom to train without licensing.
• Competitive implication: US-based AI firms (OpenAI, Meta, Anthropic, Google DeepMind) have confidence they can train on copyrighted materials under fair use and face limited legal risk. This is a material competitive advantage in cost of training and speed of model development.

UK firms do not have equivalent fair use confidence, as UK copyright law does not include a broad fair use doctrine. The decision to delay exceptions therefore leaves UK labs in legal limbo—unable to rely on statutory exceptions and vulnerable to licensing costs US competitors may not face.

Other Jurisdictions

Canada and Australia are adopting intermediate approaches, with Canada piloting a voluntary licensing framework not dissimilar to the UK model. Japan and Singapore are permitting broader AI training under research exemptions. These do not yet represent a consensus, but the trend globally is toward either statutory exceptions (EU tendency) or fair use/research exemptions (US, common law countries) rather than pure licensing reliance.

Impact on UK AI Competitiveness

Training Cost Disadvantage

Generative AI model development depends on large-scale training data acquisition. US labs benefit from fair use freedom; EU labs increasingly benefit from clearer exception frameworks. UK labs must negotiate licensing for any substantial copyrighted training data. This translates to:

• Higher upfront development costs for UK startups and mid-market AI firms
• Longer time-to-market for UK models
• Reduced competitive viability for UK open-source AI initiatives

A mid-market UK AI firm building a language model might spend £2-5 million on licensing copyrighted text; a US equivalent might spend £100-500k and rely on fair use. Over a sector, this compounds.

Talent and Investment Migration

UK AI talent—particularly researchers and engineers in foundation model development—faces incentives to relocate. The AI talent market is global. Researchers prefer working at organisations with clearer regulatory footing and lower compliance costs. Investment capital follows similar logic. VCs backing AI infrastructure and models now increasingly see the UK as higher-risk due to copyright uncertainty.

The Alan Turing Institute and UK universities remain world-class, but their ability to retain and recruit AI talent depends partly on downstream commercial viability for spin-outs and collaborations. Policy uncertainty undermines that pathway.

Export and Licensing Asymmetry

UK AI products and models trained under licensing restrictions face different cost structures than US or EU equivalents. When exporting to EU or US markets, UK developers may compete against models trained more cheaply. Conversely, UK licensors (music publishers, text publishers, visual rights owners) may find themselves in weaker negotiating positions if UK exceptions remain uncertain while EU and US frameworks crystallise.

Industry and Expert Reaction

Analysis since the 18 March announcement reveals mixed but predominantly critical sentiment:

• AI startups and scale-ups: The AI Council (an industry body chaired by enterprise AI leaders) called the decision a "missed opportunity for clarity" and warned of "competitive leakage" to US and EU competitors.
• Creative industries: The Publishers Association and BPI welcomed the delay, but acknowledged licensing frameworks remain nascent and non-binding.
• Academic and legal voices: Leading IP scholars at Oxford, Cambridge, and LSE published a joint statement arguing the delay risks entrenching licensing power imbalances (established players with licensing capacity gain advantage over innovators).
• Government justification: DSIT officials have stated privately that evidence of licensing effectiveness will inform the Q1 2027 review. The gamble is that by then, EU and US approaches will be clear enough for UK alignment.

Forward-Looking Analysis: Three Scenarios

Scenario 1: EU Precedent (Probability ~40%)

If the EU crystallises opt-out exceptions by Q4 2026, the UK government will likely align, implementing similar exceptions in 2027. This would be the least disruptive outcome for UK competitiveness but represents a one-year delay. Firms that deferred UK-based model development will restart, but some will have already shifted investment to EU jurisdictions.

Scenario 2: Licensing-Led Market (Probability ~35%)

If licensing frameworks mature and become effectively mandatory through market pressure, UK firms will operate under a de facto licensing regime while US competitors rely on fair use. This entrenches a cost disadvantage and favours UK firms with strong balance sheets over innovators. The UK's innovation ecosystem suffers incrementally.

Scenario 3: Regulatory Fragmentation (Probability ~25%)

If the EU, US, and UK end up with substantially different approaches (e.g., EU with opt-out exceptions, US with fair use, UK with licensing), global AI development fractures into regional optimisation strategies. This is least desirable for a smaller innovation economy like the UK, which benefits from unified global standards.

What Should Change: Policy Recommendations

For CAIOs and enterprise leaders seeking to influence policy, several recommendations emerge:

1. Advocate for statutory exceptions with opt-out: This balances innovation and creator compensation. Propose alignment with emerging EU frameworks.
2. Push for accelerated evidence review: The Alan Turing Institute's 12-month review is appropriate, but government should commit to action by Q3 2026 rather than Q1 2027, matching international timelines.
3. Establish licensing standards: If licensing is the interim path, industry and government should jointly establish binding standards for fair pricing and transparency, preventing incumbent lock-in.
4. Create innovation carve-outs: Consider time-limited exceptions for UK-based startups and SMEs in AI, allowing them to train freely for a defined period (e.g., first £5m revenue or first 3 years) before licensing obligations commence.

Conclusion: The Cost of Waiting

The UK's 18 March 2026 decision to shelve copyright exceptions in favour of monitoring and licensing represents a policy choice, not an inevitability. It reflects genuine tensions between innovation incentives and creator compensation, but it resolves those tensions in a way that advantages incumbents and delays certainty.

For UK AI competitiveness, the cost of waiting is measurable: delayed model development, higher training costs, talent emigration, and venture capital reallocation. These are not hypothetical. Startups are already making investment decisions based on regulatory certainty elsewhere.

The government's Q1 2027 review will be critical. By then, EU and US frameworks should be clearer. If the evidence shows licensing markets are not delivering fairness and transparency, the case for exceptions will be overwhelming. If the EU has moved decisively toward opt-out exceptions, UK alignment becomes politically easier.

The risk is that by Q1 2027, UK AI capability will already have shifted to more certain jurisdictions. Policy can be reversed; competitive advantage, once lost, is harder to reclaim.

Next reading: See our companion articles on EU AI Act compliance for UK firms and emerging AI licensing models for tactical guidance on navigating current copyright uncertainty.